Managing by Feedback
Dr. Jinci Liu
Presidential Fellow
NUS Business School
Managers give feedback, but does how they say it affect worker productivity beyond what they say? Using GitHub and LinkedIn data, I analyze over 200 million code-review messages and use large language models to distinguish feedback tone (toxicity and positivity) from informational content (constructiveness). Exploiting quasi-random reviewer assignment, I find that toxic feedback reduces workers’ subsequent code output by 43% and lowers its quality, whereas nontoxic criticism has no such detrimental effects. Positive feedback improves code quality and retention, with benefits spilling over to coworkers. Constructive feedback instead reallocates effort from new code production to revision. Feedback style varies persistently across reviewers, and these differences predict 22% of the variation in reviewer value-added. These differences may partly reflect prior experience: reviewers previously exposed to toxic feedback are more likely to provide it themselves.
